Since calculating the density of a hard disk drive is crucial for determining its storage capacity, the measurement and computation processes require a considerable amount of time. The key variables affecting density are bits per inch (BPI) and tracks per inch (TPI), where these initial values are currently defined by a fixed constant derived from known passive data. In this research, we propose using machine learning from upstream data to appropriately set the initial density values for each head on the basis of its specific characteristics. Additionally, we explore enhancing efficiency by incorporating particular data into the model. The most challenging part of this research is predicting the final state after applying the share margin process, which alters the end capability of each read head depending on the characteristics of neighboring heads. To address this, we employed a learning method that leverages data from the neighboring heads to predict the final state. The results demonstrate a significant reduction in the workload of each head’s read/write process, improving their performance, and effectively increasing the areal density of hard disk drives compared with traditional methods.